Gergely Orosz has a very interesting Twitter thread about his theory/observation that people doing CTO / Head of Engineering / VP of Engineering at startups and mid-sized companies are burning out and/or leaving. The reasons should be interesting to anyone in the tech domain.
I personally don’t know anyone like this, but Gergely gives a list of reasons (that I’ve reproduced below with my comments). They make a lot of sense to me, so my feeling is that:
It is probably happening to at least some people
It is probably happening in Silicon Valley right now, but will move to other parts of the world, including India
If at least some of the reasons don’t resonate with you, maybe you’re not paying close attention to what is happening on the cutting edge of agentic AI
Here’s Gergely’s list of reasons, and my comment after each reason. Even if you’re not a CTO or VP/Head of Engineering, understanding these reasons and mapping them to your own context might be a useful exercise.
1. Realizing the role they were hired into is just a sh*t one to do right now. Thanks to AI, often unrealistic expectations
I think there are three kinds of customers/bosses in the world right now.
Those who really understand the strengths and weaknesses of AI and/or are excellent at taking wise decisions with incomplete data, and are doing a good job of stewarding their teams. These are rare, and you’re lucky if you have one of these.
Those who don’t understand AI, but have unrealistic expectations from it, and are pushing teams—this can be a reason for the burnout that Gergely is taking about, and
Those who don’t understand AI, are ignoring it, and continuing with life as usual—this is a recipe for trouble.
Which one do you have, and how does that affect your future?
2. Realize the company they hired doesn't have a future/strategy to come out winning w AI
There are a lot of such companies right now. I don’t know if AI will be a net creator or destroyer of jobs, but I know for sure that it will cause a lot of churn in the industry, creating a lot of losers and winners.
Do you want to be stuck with a loser?
3. Realize they will not get the AI-native experience they know they need to get to thrive long-term, and the setup won't allow them to get it, so it would be career suicide
One of the biggest problems I see right now is that in most companies, company policies are hobbling their employees. If your company does not give you access to Claude Code or Codex CLI, you are at a disadvantage. If you’re personally paying for one (or both!) of them and using it for your work, you might still be OK. But if your company is actively preventing you from using those (because of concerns around privacy, company IP, or regulations), then, as Gergely says, you are not getting the AI-native experience you need to get to thrive long term.
4. The person they replaced went through some of these and it's why they left
5. Those REALLY good in metros (eg NYC, SF) can just do fractional CTO work... and so they do. Especially with all the above
6. These are folks who NEED to read the business and the room. And they realize joining the right company v important now. Also very hard
#6 is probably true for the same reasons as #2. But also, if you’re a developer and you haven’t so far developed the skills to understand the business (or to “read the room”), then again, you are at a disadvantage. Being a techie who’s not concerned with business issues (or office politics) was acceptable until yesterday, but isn’t any more. AI is commoditizing programming and tech skills, and your ability to bring important context to the AI becomes the reason you have a job. And the most important context is an understanding of the business.
7. Starting your own thing when you are a very good eng leader is easier than before. These folks can also raise funding. So some consider doing this, and eg start to build on the side
If you’re the type who wants to start a startup1, this is probably one of the best times to do it.
8. Some realize they are burnt out, and now there's a good time to take a career break of a few months
I’m not sure of this. Maybe he means that now is a good time to take a break, and seriously spend a few months becoming really good at using agentic AI.
All of the Above Might Be Wrong
Nobody understands how AI is going to change the world. So it is possible that everything I’ve said above is wrong.
If you disagree, please leave a comment explaining how and why. You might be right, and I might change my mind based on your comment2.
No. You shouldn’t start a startup. If you have to go around asking people if you should start a startup, then the answer is always No. The only people who should start a startup are those who will start one in spite of being told that they shouldn’t start a startup
This is the time to strongly believe in the “strong opinions weakly held” philosophy.
Navin
Best statement ever “ No. You shouldn’t start a startup. If you have to go around asking people if you should start a startup, then the answer is always No. The only people who should start a startup are those who will start one in spite of being told that they shouldn’t start a startup”
I will be using this to answer the countless times I have been asked this question.
I will list down what I think could be the reasons for burnout. I will categorize them into two buckets:
1. Use of AI in the software development process
* Here, either the person is swamped with pull requests that are too big to review carefully, or are AI slop, or people are not taking responsibility and saying "AI did it", or it's a combination of these.
* It could also be the pressure of proving that the team has become more productive, or something of that sort.
In my personal experience, developers are actually the first and biggest adopters of AI. However, in the overall SDLC, developers are only one part of the process. Customers, product managers, QA, etc. have been much slower to adopt AI tools in their own workflows, so there isn't much to show in terms of overall productivity gains.
2. Use of AI in the product itself
* Here, the challenge could be using a Voice AI agent, or a customer aggressively asking to slap AI into the product somewhere.
* Or trying to prove or QA that an inherently non-deterministic system behaves exactly the same way every time.
But I don't agree with some of the "missing out on AI" arguments:
* If this is about actually using machine learning tools and algorithms, such as PyTorch, TensorFlow, or the ability to train a small model, then I get it.
* But for the majority of users, it's really about "using" or "interfacing" with AI models. In that case, you can learn and experiment at home; you don't have to depend on the company's work or policies.